Vehicle Path Reconstruction Using Automatic Vehicle Identification Data: A Bi-Directional Long Short-Term Memory-Based Approach

Vehicle path reconstruction plays an important role in transportation field, which helps estimate traffic flow and predict traffic congestion, etc. However, existing methods resort to either behavior modeling based on user equilibrium and route choice assumption or learning approaches heavily relying on prior estimation. Based on Automatic Vehicle Identification (AVI) data in Baoding City, Hebei Province, China, this study uses a Bi-directional Long Short-Term Memory (Bi-LSTM) approach to reconstruct the incomplete vehicle path. By extracting the complete vehicle path using AVI data, the pre-processed vehicle paths are converted to a series of path strings. By implementing the one-hot encoding, the path strings are then converted to vectors, which are trained as the input of Bi-LSTM network. Then Bi-LSTM model is used to learn the whole information of input data and predict the vehicle identification data that needs to be reconstructed. The results show that the proposed approach achieves a high accuracy of 93.56% and performs better than dynamic RNN and shortest-path.